Moving from OrbStack to Apple Container, with PostgreSQL in Tow

I am not a native English speaker; this article was translated by AI.

Today I removed OrbStack from my Mac.

Not because it was bad. Quite the opposite: for the past few years, OrbStack has been the least annoying Docker replacement I have used on macOS. But Apple’s own container CLI can now run Linux containers, and the only long-running local service I still had was PostgreSQL. With the dependency surface already this small, it made sense to shrink the runtime too.

The dangerous version of this migration would be: “just delete it and recreate it.” Containers can be recreated. Databases cannot.

First, check what is actually running

#

There was only one long-running container in OrbStack:

docker ps -a --format '{{.ID}}\t{{.Image}}\t{{.Names}}\t{{.Status}}\t{{.Ports}}'

The result was simple:

d377e0104620  registry.cheftin.cn/p/postgres17_pd-arm64  scriber_pg  Up 2 weeks (healthy)

Then docker inspect exposed the trap: docker-compose.yml mounted ./pg17:/var/lib/postgresql/data, but the image actually used PGDATA=/data/data. In other words, the host-side pg17 directory was empty, while the real database lived inside the old container under /data.

That almost turned the migration into an incident.

PGDATA=/data/data
/Users/ferstar/myprojects/postgresql/pg17  0B
/data                                 7.2G

So I could not simply stop OrbStack and remount the same host directory. I had to export the data from the running old container first.

Backups first

#

I made two backups:

docker exec scriber_pg pg_dumpall -U postgres | gzip -9 > pg_dumpall.sql.gz
docker exec scriber_pg tar -C / -cf - data | zstd -T0 -10 > container-data.tar.zst

The logical backup was only 38M; the physical archive was 6.8G. At first I thought the database was huge. It was not.

The largest part of /data was not business data, but the image’s own local WAL archive:

6865.7M  data/backup/wal
293.5M   data/data/base
126.7M   data/data/log
32.0M    data/data/pg_wal

The actual table data was only a few hundred MB.

Apple Container takes over

#

This machine is on macOS 15.7.3. Apple’s container project is more officially aimed at macOS 26, but Homebrew can already install it:

brew install container
container system start

The first startup installs the default kernel. After a small Alpine test passed, I exported the PostgreSQL image from local Docker and loaded it into Apple Container, avoiding private registry credential issues:

docker save registry.cheftin.cn/p/postgres17_pd-arm64:latest -o postgres17_pd-arm64.docker-save.tar
container image load -i postgres17_pd-arm64.docker-save.tar

The final container is launched like this:

container run -d \
  --name scriber_pg \
  --memory 4g \
  --cpus 4 \
  --shm-size 8g \
  -p 0.0.0.0:5432:5432 \
  -v /Users/ferstar/myprojects/postgresql/apple-pg17:/data \
  -e DEBUG=false \
  -e POSTGRES_PASSWORD=... \
  -e POSTGRES_HOST_AUTH_METHOD=trust \
  registry.cheftin.cn/p/postgres17_pd-arm64:latest \
  postgres -c archive_mode=off -c archive_command=

The last two -c options matter.

This image rewrites postgresql.conf during startup, so editing the file directly inside the data directory gets overwritten. Putting archive_mode=off in PostgreSQL command-line arguments makes the setting source command line, and it stays stable after restart.

Verification:

archive_command=(disabled) source=command line
archive_mode=off source=command line
wal_level=logical source=configuration file

I left wal_level=logical alone. It was not the source of the local backup growth, and disabling it casually could break logical replication or extensions.

One more piece for autostart

#

Apple Container does not yet feel like Docker Compose with restart: unless-stopped. brew services start container brings up the apiserver at login, but it does not guarantee a specific container starts.

So I added a user LaunchAgent:

~/Library/LaunchAgents/com.ferstar.apple-container.scriber-pg.plist

It runs a tiny script:

container system start >/tmp/scriber_pg_container_system_start.log 2>&1 || true

if container ls --quiet | grep -Fxq 'scriber_pg'; then
  exit 0
fi

container start scriber_pg

Not elegant, but enough.

The final bill

#

Then OrbStack was uninstalled and zapped:

brew uninstall --cask --zap orbstack

I also removed leftovers under ~/.orbstack, ~/OrbStack, and related Group Containers directories. Since the local Docker CLI came from OrbStack, the docker command disappeared too. That is fine; daily usage now only needs container.

Cleanup result:

Free space at the tightest point:  6.8GiB
After deleting physical rollback archive: 14GiB
After uninstalling OrbStack:       26GiB
Total freed:                       about 19GiB

Current footprint:

Apple Container runtime data: 3.8G
New PostgreSQL data directory: 274M
Migration logical backups/logs: 76M
scriber_pg actual memory: about 148MiB / 4GiB
scriber_pg CPU: 0.00%

The databases are still there:

postgres|11

Summary

#

The biggest lesson here was not how to use Apple Container. It was: do not trust the volume path in a compose file at face value. Check the running container’s real PGDATA and mounts first.

The other lesson was to distinguish between WAL that PostgreSQL needs to run and WAL that an image archives elsewhere as backup. The former must stay. The latter should be disabled if local backups are not needed. Otherwise, a few-hundred-MB database can eventually scare you with several GB of archived WAL.

OrbStack is gone. PostgreSQL stays. One fewer always-on runtime for this Mac.